Influence of surface roughness on measurement uncertainty in Computed Tomography

Transcription

1 Influence of surface roughness on measurement uncertainty in Computed Tomography More info about this article: Leonard Schild, Alexandra Kraemer, Doreen Reiling, Hanjue Wu, Gisela Lanza wbk Institute of Production Science, Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany, Abstract In geometrical measurements using computed tomography (CT), the measurement uncertainty is often derived from reference measurements that have been made with a tactile coordinate measuring machine (CMM). This paper shows the influence that a test object s surface roughness has on such reference measurements. To do so, test objects which have been manufactured by means of additive manufacturing methods are compared to conventionally produced objects. Results of geometrical measurements using a tactile CMM as well as measurements conducted with a CT are analysed with respect to the surface quality. Following the analysis, the validity of tactile CMMs to reference CT measurements is discussed. Keywords: dimensional metrology, surface quality, measurement uncertainty, reference measurement, additive manufacturing 1 Introduction X-ray computed tomography (CT) has already been used in dimensional metrology for the last ten years. Nevertheless, traceability for the purpose of estimating the measurement uncertainty is still one of the technology s weaknesses in comparison with tactile coordinate measuring machines (CMM) [1]. VDI [2] proposes reference measurements as a state-of-the-art method to obtain the uncertainty of CT measurements. This leads to the need for a valid reference measuring technique. VDI suggests using a CMM to conduct the reference measurements. Consequently, the uncertainty of the CMM measurement becomes part of the measurement uncertainty of the CT. This means the CTs uncertainty can be estimated properly only if accurate CMM measurements are available. For parts manufactured by conventional means of production (e.g. turning, milling) this requirement can be fulfilled. However, modern additive manufacturing methods such as selective laser beam melting (SLM) lead to a poor surface quality. This lack of a smooth surface imposes difficulties for high precision tactile measurements and puts CMMs in question, when being deployed as reference measuring system for CT measurements. In contrast though, such additively manufactured parts are suitable for being investigated with CT. They often comprise geometric features that cannot be measured with tactile CMMs but only with CT [3]. This leads to two questions: Does the surface quality of a measuring object affect the results of dimensional measurements using CT? Can CMMs be applied to assess the measurement uncertainty of CT when investigating parts with a rough surface such as parts produced by means of additive manufacturing? To answer these questions, this paper proposes to design suitable test objects and presents corresponding experiments using such test objects. 2 Influence of surface roughness in CT According to ISO [4], the extended measurement uncertainty for a measurement process using a calibrated work piece 2 is calculated by the formula = In this formula, represents the uncertainty from the calibration, from the measurement process, from the work piece and the bias of the measurement process. This concept is used by many authors to estimate the measurement uncertainty of a CT system [5 8]. As covers the influence of the work piece s surface roughness, Schmitt et al. propose to calculate by including the roughness parameter /2 as a factor [6]. Others follow similar approaches. Bartscher et al. undertake experimental trials from which they find that the uncertainty budget caused by the roughness is roughly /4 [9]. The effect of the surface roughness on CT measurements is the topic of recent studies. Novak et al. use work pieces which have been calibrated using tactile CMM. They describe that the surface roughness and the resulting measurement uncertainty are correlated [7]. Aloisi et al. undertake experiments to verify the influence of the surface roughness on CT measurements by using calibrated work pieces as well [8]. The work pieces are produced by additive manufacturing methods and calibrated by tactile CMM. It is shown that the deviation between tactile CMM and CT measurements are of roughly the same magnitude as the surface roughness. Carmignato et al. undertake like-minded experiments. By also using calibrated work pieces that have been produced by additive means, they show that the surface roughness causes uncertainty of the same magnitude as the value of the surface roughness [10]. 1

2 In summary, the influence of the surface roughness of the work piece on the measurement uncertainty is a well-studied topic. All above mentioned pieces of work use tactile calibration. However, the influence different parameters of the CMM like the probe diameter have on the calibration s uncertainty and the consequences they have on the estimation of the CTs uncertainty are not discussed. Additionally, it does not get talked about the possibility of a rough surface being in need of different CT settings compared to a smooth surface in order to achieve a low measurement uncertainty. 3 Design of test bodies Test bodies that enable the questions above to be answered need to fulfil some requirements. The most basic ones are: It must be possible to conduct dimensional measurements with the test objects using a CMM as well as a CT. The test objects must allow a comparison between different surface qualities. Additionally, a few optional requirements are derived from practical needs: To save time, the test objects should allow automated measurements and should not comprise too many features. To detect influences imposed by CT thresholds, uni- as well as bi-directional measurements should be possible with the test objects. To broaden the test objects scope, their form may represent parts that are in industrial use. According to these criteria, a test object is designed. It is manufactured using traditional methods as well as additive means. The usage of these different production methods results in test bodies with different surface qualities. Both specimen are shown in Figure 1. Figure 1. Test objects made by different manufacturing techniques (left conventional means of production, right additive) and used to assess the measurement uncertainty of a CT system. VDI [11] recommends using a stepped cylinder in order to determine the measurement uncertainty of a CT system. It represents industrial use cases and fulfils most criteria mentioned above. As it is made of a number of steps with varying thickness, the radiation length s influence on the uncertainty can be estimated. Kasperl et al. suggest an asymmetrical stepped wedge as reference object in order to additionally estimate the influence of symmetry [12]. However, neither of these test bodies has directly comparable uni- nor bidirectional measurement features. Therefore, a new test body is designed based on the stepped wedge. It is basically a stepped wedge with additional holes (see technical drawing in Figure 2). These permit comparisons between two bidirectional measures (diameters and bi-directional distances). Additionally this work piece has two unidirectional measures (diameter distances and uni-directional distances), which can also be compared. This test body is manufactured using milling and selective laser beam melting (SLM) (see Figure 1). No. of test feature in Type of measurement drawing feature 1 2 Diameter Bi-directional distance Uni-directional distance Distance of diameters 14 Figure 2. Technical drawing of test object including measurement features. 2

3 4 Experiments After designing and manufacturing the test objects, different measurements are undertaken using a CMM as well as a CT. To answer the questions from the beginning, parameters of the measuring systems are varied in the course of different experiments (see Table 1). CMM Experiments As stated in the state-of-the-art section, a referencing method for calibrating the reference workpieces is needed in order to estimate the measurement uncertainty of CT measurements. This reference method should preferably be traceable tactile CMM. The CMM Zeiss O-Inspect 322 (MPE 3.2 µm) is used for the additively manufactured work piece as well as for the conventionally manufactured workpiece. But, especially with rough surfaces, tactile CMMs suffer from a mechanical filter effect [13]. This is due to the surface roughness and the diameter of the CMM probe. A probe with a large diameter, in contrast to a smaller one, is not able to reach the valleys of a rough surface. This leads to the recorded points varying, depending on which probe has been used. As the large probe is only able to detect big differences in local surface roughness, a large probe acts as a low pass filter. This behaviour leads to best-fit elements like circles varying in size and positon. For example, the diameter of a drill is measured with a smaller value when a big probe is used. As the surface of an additively manufactured workpiece is very rough, the influence of the probe diameter on the measurements needs to be analysed. Additionally, preliminary experiments show that the measurement force used also has an effect on the recorded points. An explanation for this observation is, that the probe slips into valleys on the surface easily if a large measurement force is used. To show these effects and quantify their influence, three different experiments are undertaken. Experiment 1: small probe diameter (0.9mm), small measurement force (15mN) Experiment 2: small probe diameter (0.9mm), large measurement force (30mN) Experiment 3: large probe diameter (1.5mm), small measurement force (15mN) All experiments have been repeated five times. The measurement plan and the clamping arrangement (see Figure 3) have been set up identically in all of the 15 executions. In the figure showing the clamping arrangement it can be seen that the surface of the work piece is extremely rough. As the roughness varies over the surface, no measurement was carried out. However, as the roughness is palpable, it is estimated to lead to an Rz-value in the magnitude of several tenths of a millimetre. Figure 3. Clamping of additively manufactured work piece in CMM The results of the experiments are shown in Table 1. In this table the average values of the repeated experiments are displayed for all measures. Additionally, the average deviation from the technical drawings is supplied in order to allow conclusions on how well the SLM process has worked. Furthermore, a double standard deviation for all measures is provided in order to assess the variance of the experiment repetitions. It can be seen, that the average deviation from the drawings varies in all three experiments, as expected, but only measurements undertaken using the large probe have small random deviations. An explanation for this behaviour could be that even with a small probe force, the small probe will slip into different valleys on the surface of the part if the clamping position differs only slightly. Another explanation is that the shaft of the probe touches the work piece and thus causes a deviation because illegal points are recorded. As the probe diameter is only very small differences in form may cause this behaviour in the case of the smaller probe. Nevertheless, a bigger probe should be used to achieve repeatable results. However, this means applying a mechanical low-pass filter. As explained above, this low pass filter behaviour causes a large systematic deviation. This poses the question whether a CMM is really suitable for calibrating work pieces with a rough surface. For this reason, a measurement uncertainty estimation is not presented. The experiments have shown that this CMM is not traceable for this measurement task. For the sake of comparison, the mean values of the third experiment (large probe) are used as calibration values when analysing the CT measurements. In addition to the results for the additively manufactured work piece, the results for the tactile calibration of the conventionally manufactured work piece are shown in Table 1. The tactile calibration for this work piece has been executed on the same CMM using the same 1.5mm probe as used for the additively manufactured workpiece, but, the experiments have been undertaking using a measurement force of 75mN and have been repeated twenty times. Looking at the results from the conventionally produced part, the random deviation is one magnitude lower than that of the experiments with the additively manufactured work piece. This shows the huge impact the surface roughness has on tactile CMM measurement. As a mechanical filter impact cannot be expected, these results could be used for a measurement uncertainty calculation. 3

4 Table 1. Results of tactile calibration. On display are results for the additively manufactured test object (columns with abbreviation Add.) as well as the results for the conventional test object (columns with abbreviation Con.). For the additively manufactured test object all three settings are shown (columns with abbreviation Sett. 1, etc.). The first setting is small probe diameter and low force, the second is small probe diameter and high force while the third is big probe and low force. Average deviation from technical Average values in mm Measurement drawing in m 2x standard deviation in m feature Add., Add., Add., Add., Add., Add., Add., Add., Add., Con. Con. Sett. 1 Sett. 2 Sett. 3 Sett. 1 Sett. 2 Sett. 3 Sett. 1 Sett. 2 Sett. 3 Con. Diameter 1 4,949 4,931 4,981 4,996 50,8 68,8 19,3 3,6 40,7 81,5 2,7 0,3 Diameter 2 4,950 4,930 4,988 4,998 49,5 69,7 11,8 2,5 41,7 87,4 4,3 0,2 Diameter 3 4,940 4,919 4,980 5,000 60,4 80,6 19,7 0,4 36,3 85,2 4,3 0,2 Diameter 4 4,958 4,936 4,979 5,000 42,1 63,8 21,4-0,2 37,1 85,9 4,9 0,2 Bi-direct. distance 1 7,934 7,959 7,925 7,953 65,8 41,1 74,9 47,2 24,1 100,2 2,5 1,4 Bi-direct. distance 2 9,934 9,960 9,921 9,962 65,6 40,0 78,7 38,1 33,4 103,4 1,3 1,7 Bi-direct. distance 3 11,914 11,927 11,914 11,970 86,3 72,8 85,5 29,7 10,5 58,7 3,1 1,4 Bi-direct. distance 4 13,912 13,934 13,904 13,981 88,1 65,6 96,0 18,7 23,4 90,8 2,2 2,1 Uni-direct. distance 1 2,016 2,010 2,005 2,001-16,1-9,5-4,8-1,3 7,1 14,6 5,7 1,2 Uni-direct. distance 2 4,041 4,024 4,018 4,000-41,2-23,7-17,8-0,2 25,9 38,5 13,3 2,4 Uni-direct. distance 3 6,072 6,049 6,031 6,000-72,3-49,0-31,4 0,1 36,9 52,4 19,5 3,4 Diameter distance 1 7,987 7,988 7,970 8,001 13,1 11,6 29,6-1,3 2,2 4,9 3,5 0,3 Diameter distance 2 15,957 15,958 15,949 16,000 42,9 41,8 51,5-0,4 5,7 8,3 2,8 0,7 Diameter distance 3 23,940 23,942 23,926 24,002 60,0 57,6 73,9-2,2 10,2 11,8 2,2 0,8 CT experiments In the following section the results of different CT experiments are presented (see Figure 4 Figure 11). All experiments aim to compare the measurements obtained with the tactile CMM with those from the CT. To do so, the same measurement features of the same workpieces have been measured with varying settings. Afterwards the deviation, which is depicted in the following figures, between the CMM measurements and the CT measurements has been calculated. To do so, the difference between the CT measurements and the average values from Table 1 were calculated for every experiment and every feature. By doing so, the average CMM values act as calibration for the CT. Because the results from the third CMM experiment on the additively manufactured work piece have the lowest random deviation, these are used as the calibration values for the additively manufactured work piece. Besides just defining the differences between CMM and CT measurements, the experiments also aim to show the influence various CT settings have on the deviation, in order to assess which setting causes the smallest deviation. To achieve this target, a design of experiments (DOE) has been used. All experiments were performed on a Zeiss Metrotom 800 in combination with VG Studio 2.2. The experimental design consists of five parameters in two steps, which have been varied in a partial factorial design. They are presented in Table 2. Table 2. Factors used in the DOE for the CT experiments. Factor Setting low (-) Setting high (+) Means of manufacturing Additive Conventional Tube voltage 80V 100V Tube current 150mA 210mA Detector gain 1x 2.5x Pre-Filter no filter 0.25 mm Cu The experimental design would consist of 32 experiments if it were full factorial. To halve the number of experiments, only half of the other settings has been carried out in combination with a pre filter. Table 3 shows the experimental design for all factors except for the means of manufacturing, as the same experiments have been conducted for both means of manufacturing. In order to keep all experiments comparable, they have been performed using the same integration time and the same distance from object to source. Additionally, the same script has been used in VG studio to extract the measurement features automatically and the surface determination was set to automatic. Table 3. Experimental design of the DOE used for the CT experiments. Parameter Set Voltage Current Gain Filter The following figures show the results from the DOE. All figures have the same scale so as to be easily comparable. Each figure shows the deviations between the CT measurements and the calibration results provided in Table 1. Details on which measurement feature is which are provide by the technical drawings in Figure 2. The figures feature error bars which represent 4

5 two times the standard deviation. All experiment have been carried out three times to assess the random deviation. For the ordering of the figures, first the results for the additively manufactured part are shown, then for the conventionally manufactured. Each group of measurement features is depicted in a separate figure which contains the results for all settings from the DOE. Figure 4 shows the deviation of the diameters and Figure 5 of the bi-directional distance for the additively manufactured part. Both measurement feature groups are bi-directional and yield similar results: parameter sets 1-4 entail differences in the systematic deviation compared to settings 5-8, while random deviations for all sets are similar in magnitude. Thus, two groups of parameters exist. The first group consists of parameters without filter (set 1-4), while the second group consists only of settings that include the use of a filter (set 5-8). Looking closely at these two groups, it can be seen that the systematic deviations within these parameter set groups are roughly equal for all settings and all diameters. The same observation can also be made for all bidirectional distances. The systematic deviations vary between the diameters and distances respectively for sets 1-4 much more than for sets 5-8. Additionally, random deviation is also bigger for settings 1-4. Otherwise, the results in the two groups of settings are fairly similar. A very close inspection of the graphs may show that the variation in the systematic deviations is smaller between settings in the filter-usage group compared to the group using no filter. However, this observation may be related to the small of number of experiment repetitions. Figure 6 and Figure 7 show results for the unidirectional measures (uni-directional distance and diameter distances). In these measurements no clear groups of settings can be spotted: all settings have random and systematic deviations of similar magnitude. Closer interpretation is hindered by the small number of experiment repetitions. Figure 4. Results for CT experiments on the diameters for the additively manufactured work piece. Figure 5. Results for CT experiments on the bi-directional distances for the additively manufactured work piece. 5

6 Figure 6. Results for CT experiments on the uni-directional distance for the additively manufactured work piece. Figure 7. Results for CT experiments on the diameter distances for the additively manufactured work piece. Figure 8 Figure 11 show the results for the conventionally manufactured parts. There are similarities to the results of the additively manufactured work piece: while bi-directional measurement features lead to two groups of parameters sets, those with and those without a filter, this behaviour cannot be seen with the uni-directional measurement features. But, in contrast to the results from the additively manufactured workpiece, the results with the conventionally manufactured work piece have much smaller random deviations. This holds for all measurement feature groups except for the diameter distances. Additionally, the systematic deviations in the with-filter group of settings are more homogeneous than those in the without-filter group for the bidirectional measurement features (compare Figure 8, Figure 9). So, for the conventionally manufactured work piece a pre-filter should be used. An explanation as to why the results for the diameter distances (Figure 11) are so much worse in terms of random deviation compared to the uni-directional distance, can only be guessed at. To calculate the distance, a cylinder is fitted into each bore hole. Afterwards, the centre of each best-fit cylinder is projected onto the plane on top of the workpiece. Then, the distances of the projected centres are measured. Therefore, there are many aspects that may vary, such as the position of the cylinder s axis, and these could have a big influence on the measurements and make them prone to big random deviations. 6

7 Figure 8. Results for CT experiments on the diameters for the conventionally manufactured work piece. Figure 9. Results for CT experiments on the bi-directional distances for the conventionally manufactured work piece. Figure 10. Results for CT experiments on the uni-directional distances for the conventionally manufactured work piece. 7

8 Figure 11. Results for CT experiments on the diameter distance for the conventionally manufactured work piece. Summary and Outlook The results presented in this paper show that a tactile CMM is not ideal for calibrating additively manufactured work pieces which have a rough surface. To obtain good results regarding repeatability, a probe with a large diameter needs to be used. However, as a large probe diameter acts as a mechanical low pass filter, a systematic deviation is to be expected for such measurements. Thus, the CMMs measurement uncertainty get very large. This leads to it not being well-suited as a reference for less accurate measurement technologies such as CT. Furthermore, some recommendations can be derived from the CT experiments presented in this paper. The experiments have shown that similar observations can be made when measuring additively and conventionally manufactured work pieces. First, a pre-filter causes results for bi-directional measurement features to differ greatly from results obtained without using a pre-filter. Other settings, like the tube voltage or current do not have a significant influence on the systematic or the random deviation from a calibration value. The random deviation with additively manufactured work pieces is bigger, however, than that of conventionally manufactured work pieces. In order to assess the systematic measurement uncertainty of CT, a well-calibrated work piece is needed. But, as the experiments on the CMM have shown, a CMM is not able to provide these for work pieces with a rough surface. To compensate for this shortcoming, rules need to be defined which help to make the measured dimensions of additively manufactured work pieces comparable. For example, the CMMs probe diameter could be set as a function of the surface roughness value Rz. Alternatively, other reference systems for CT need to be found in order to be able to assess the measurement uncertainty of CT for additively manufactured work pieces. The research has been supported by the research project LA2351/37-1 Optimierung von Aufnahmeparametern mittels projektionsbasierter Qualitätskenngrößen in der industriellen Computertomographie of the German Research foundation. This is gratefully acknowledged. References [1] P. Müller, J. Hiller, A. Cantatore, and L. de Chiffre, A study on evaluation strategies in dimensional X-ray computed tomography by estimation of measurement uncertainties, Int. J. Metrol. Qual. Eng., vol. 3, no. 2, pp , [2] Computertomografie in der dimensionellen Messtechnik - Bestimmung der Messunsicherheit und der Prüfprozesseignung von Koordinatenmessgeräten mit CT-Sensoren, VDI/VDE , [3] L. De Chiffre, S. Carmignato, J.-P. Kruth, R. Schmitt, and A. Weckenmann, Industrial applications of computed tomography, CIRP Annals - Manufacturing Technology, vol. 63, no. 2, pp , [4] Geometrische Produktspezifikation und -prüfung (GPS) Verfahren zur Ermittlung der Messunsicherheit von Koordinatenmessgeräten (KMG) Teil 3: Anwendung von kalibrierten Werkstücken oder Normalen, DIN EN ISO , [5] C. Affenzeller, C. Gusenbauer, M. Reiter, and J. Kastner, Measurement uncertainty evaluation of an X-ray computed tomography system, in International Symposium on Digital Industrial Radiology and Computed Tomography (DIR 2015), Ghent, [6] R. Schmitt and C. Niggemann, Uncertainty in measurement for x-ray-computed tomography using calibrated work pieces, Measurement Science and Technology, vol. 21, no. 5, p , [7] H. Novak, Amalija, Runje, and Biserka, Influence of object surface roughness in CT dimensional measurements, in 7th International Conference on Industrial Computed Tomography (ict) 2017: 7-9 Feb, Leuven, Belgium. [8] V. Aloisi and S. Carmignato, Influence of surface roughness on X-ray computed tomography dimensional measurements of additive manufactured parts, Case Studies in Nondestructive Testing and Evaluation, [9] M. Bartscher, M. Neukamm, M. Koch, and et al., Performance Assessment of Geometry Measurements with Micro-CT Using a Dismountable Work-Piece-Near Reference Standard, in 10th European Conference on Non-destructive Testing 2010, Northampton, UK, [10] S. Carmignato, V. Aloisi, F. Medeossi, F. Zanini, and E. Savio, Influence of surface roughness on computed tomography dimensional measurements, CIRP Annals, vol. 66, no. 1, pp , [11] VDI/VDE 2630 Blatt 1.3: Computertomografie in der dimensionellen Messtechnik - Leitfaden zur Anwendung von DIN EN ISO für Koordinatenmessgeräte mit CT- Sensoren, VDI/VDE , [12] S. Kasperl and J. Hiller, Artefaktkorrekturen beim dimensionellen Messen mit industrieller Röntgen-Computertomographie, Technisches Messen, vol. 76, no. 9, pp , [13] A. Weckenmann, T. Estler, G. Peggs, and D. McMurtry, Probing Systems in Dimensional Metrology, CIRP Annals, vol. 53, no. 2, pp ,